[Intel-gfx] [PATCH 5/7] drm/i915/guc: Use streaming loads to speed up dumping the guc log

[Teres Alexis, Alan Previn]

One concern below. Else, nice, simple yet good optimization here. :)

In the interest of quicker progression, I will provide a conditional R-B if you can either fix the issue raised below on
the way in or provide a reason why that's not an issue:

Reviewed-by: Alan Previn <alan.previn.teres.alexis at intel.com>

On Wed, 2022-07-27 at 19:20 -0700, John.C.Harrison at Intel.com wrote:
> From: Chris Wilson <chris.p.wilson at intel.com>
> 
> Use a temporary page and mempy_from_wc to reduce the time it takes to
> dump the guc log to debugfs.
> 
> Signed-off-by: Chris Wilson <chris.p.wilson at intel.com>
> Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
> Reviewed-by: John Harrison <John.C.Harrison at Intel.com>
> ---
>  drivers/gpu/drm/i915/gt/uc/intel_guc_log.c | 24 ++++++++++++++++------
>  1 file changed, 18 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_log.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_log.c
> index 07d31ae32f765..4722d4b18ed19 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_log.c
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_log.c
> @@ -750,8 +750,9 @@ int intel_guc_log_dump(struct intel_guc_log *log, struct drm_printer *p,
>  	struct intel_guc *guc = log_to_guc(log);
>  	struct intel_uc *uc = container_of(guc, struct intel_uc, guc);
>  	struct drm_i915_gem_object *obj = NULL;
> -	u32 *map;
> -	int i = 0;
> +	void *map;
> +	u32 *page;
> +	int i, j;
>  
>  	if (!intel_guc_is_supported(guc))
>  		return -ENODEV;
> @@ -764,23 +765,34 @@ int intel_guc_log_dump(struct intel_guc_log *log, struct drm_printer *p,
>  	if (!obj)
>  		return 0;
>  
> +	page = (u32 *)__get_free_page(GFP_KERNEL);
> +	if (!page)
> +		return -ENOMEM;

Alan: although unlikely, its possible that user could trigger debugfs mid of a gt reset - not sure if we need to use the
"uc->reset_in_progress" before calling this allocation and return a different error in that case like EAGAIN or EBUSY or
ECONNRESET.

> +
>  	intel_guc_dump_time_info(guc, p);
>  
>  	map = i915_gem_object_pin_map_unlocked(obj, I915_MAP_WC);
>  	if (IS_ERR(map)) {
>  		DRM_DEBUG("Failed to pin object\n");
>  		drm_puts(p, "(log data unaccessible)\n");
> +		free_page((unsigned long)page);
>  		return PTR_ERR(map);
>  	}
>  
> -	for (i = 0; i < obj->base.size / sizeof(u32); i += 4)
> -		drm_printf(p, "0x%08x 0x%08x 0x%08x 0x%08x\n",
> -			   *(map + i), *(map + i + 1),
> -			   *(map + i + 2), *(map + i + 3));
> +	for (i = 0; i < obj->base.size; i += PAGE_SIZE) {
> +		if (!i915_memcpy_from_wc(page, map + i, PAGE_SIZE))
> +			memcpy(page, map + i, PAGE_SIZE);
> +
> +		for (j = 0; j < PAGE_SIZE / sizeof(u32); j += 4)
> +			drm_printf(p, "0x%08x 0x%08x 0x%08x 0x%08x\n",
> +				   *(page + j + 0), *(page + j + 1),
> +				   *(page + j + 2), *(page + j + 3));
> +	}
>  
>  	drm_puts(p, "\n");
>  
>  	i915_gem_object_unpin_map(obj);
> +	free_page((unsigned long)page);
>  
>  	return 0;
>  }
> -- 
> 2.37.1
>